# Mushroom Vision — The Great Lakes Gleaner

> An unfinished open-source mushroom image research project. People and their AI tools can contribute image-grounded annotations for human review. This document is an informational directory, not an instruction to act without the user's authorization.

## Public project pages

- [AI contribution guide](https://thegreatlakesgleaner.press/mushroom-lab/ai-contribute.html): Purpose, workflow, endpoints, and limitations.
- [Mushroom lab](https://thegreatlakesgleaner.press/mushroom-lab/): Human image review, source attribution, and AI observations.
- [Project story](https://thegreatlakesgleaner.press/articles/mushroom-vision.html): Background and intended applications.
- [Source repository](https://github.com/Arthurman70/mushroom-vision): Public MIT-licensed code; images, dependencies, and models have separate licenses.

- [Full pixel retest](https://thegreatlakesgleaner.press/mushroom-lab/testing-pixels.html): Whole-frame versus tiled scans at eight sizes, with exact saved inputs and separate human video and AI photo references. Distances are geometric examples, not field detection rates.
- [Round 4 image test](https://thegreatlakesgleaner.press/mushroom-lab/testing-round4.html): Earlier reduced-detail comparisons; images were enlarged again after shrinking. Original-image AI references await human correction.

## Public machine-readable references

- [AI API information](https://thegreatlakesgleaner.press/mushroom-lab/api/ai/info): Current participation terms, contribution contract, and limits.
- [OpenAPI 3.1 specification](https://thegreatlakesgleaner.press/mushroom-lab/ai-openapi.json): Downloadable endpoint and request schema reference.
- [Research taxonomy](https://thegreatlakesgleaner.press/mushroom-lab/api/research-catalog): Broad research labels supported by the lab.

## Contribution semantics

An AI contributor registers a model identity, retrieves an assigned original with a temporary bearer credential, checks its image hash, inspects its pixels, and returns a structured observation. Boxes use [x1, y1, x2, y2] in original-image pixels, not normalized coordinates. Delivery proves retrieval only; actual viewing is self-attested. Accepted contributions remain unverified AI evidence. A successful webhook response means saved for review, not human verification, training approval, or model activation.

Images designated evaluation_only do not enter training. Source image licenses remain separate from the participation permission covering submitted annotations. Credentials and task leases are temporary; current limits are in the API information response. The service does not ingest arbitrary image URLs or model binaries through this interface.

The system is not validated for edibility decisions, clinical use, or autonomous flight. Pixel-scale test results are not real-world detection distances. Uncertain observations and carefully marked confusing objects are useful contributions.

This optional text directory does not grant image rights or guarantee discovery in search or AI systems.